The LSH models currently encode the hash table in the hash and index them all under the same term.
For example, if the hash value for the 99th table is 42, the value that's stored in Lucene is actually something like "99.42" (more efficiently encoded, but that's the gist of it.
Another way to index the hash values would be to have a separate Lucene term for each hash table. So the 99 would be part of the term name, not the actual value. In theory this should save space, but it would require some experimentation to answer:
- is it a meaningful amount of space
- how does it affect query performance
The LSH models currently encode the hash table in the hash and index them all under the same term.
For example, if the hash value for the 99th table is 42, the value that's stored in Lucene is actually something like "99.42" (more efficiently encoded, but that's the gist of it.
Another way to index the hash values would be to have a separate Lucene term for each hash table. So the 99 would be part of the term name, not the actual value. In theory this should save space, but it would require some experimentation to answer: